Conditional Select

_ConditionalSelect is ComfyUI's version of a ternary: result = value_if_true if condition else value_if_false. That's the whole node - three inputs, one output. What makes it worth a page is the detail most people never notice: it's lazy. Only the branch you actually pick gets evaluated. That one property is why it can safely sit inside a loop's control flow, and it's genuinely handy outside loops too.

What it does

You give it a condition (boolean), a value_if_true, and a value_if_false, and you get the selected value out. Nothing surprising. The lazy part is the trick. ComfyUI normally resolves every input before a node runs - which means if your value_if_false is the result of a long expensive chain, you pay for it even when the condition is true. _ConditionalSelect tells ComfyUI's execution engine to skip that chain when it isn't needed: it requests only the input it's actually going to return, so the unselected branch never runs at all.

In a loop that matters twice over. First, it's what makes the early-stop mechanism cheap: TensorLoopClose builds a "stop or continue?" condition and selects between an exit path and a continue path - the path not taken isn't computed. Second, it prevents whole classes of "this node got None and crashed" errors, because the branch that would produce a useless value simply isn't executed.

Inputs and outputs

• condition (boolean) - which branch to take.
• value_if_true (any) - returned when condition is true.
• value_if_false (any) - returned when condition is false.
• result - whatever you selected.

Any type works - tensors, latents, ints, strings, even the loop-state dicts this pack passes around. That's why the loop machinery uses it for everything from the bypass path (iterations = 0 → just pass through) to the "no progress, stop the loop" bail-out.

Install

Same as every node in the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/kijai/ComfyUI-NativeLooping_testing

Restart ComfyUI, or search "NativeLooping" in Manager. Dev-only, so it's hidden from the menu - you'd have to add it from a workflow that already contains it.

If you want this pattern in your own graphs, you don't need to be inside a loop to appreciate it: any workflow with a "do the expensive thing or skip it" decision can use lazy selection to cut wasted compute. Just remember the whole pack is a _testing repo from kijai - the candidate for native ComfyUI loop nodes - so don't build long-lived workflows on these internals. And one gotcha specific to lazy nodes: because the unselected input genuinely doesn't get computed, you can't count on side effects like progress bars firing on the skipped branch. Pick lazy selection for value, not for logging.